inductor: support linear fusion when multi linear using same input (#106300)
For ```llama``` model, there has a pattern that multi linear using the same input and input dim > 2:
```input->view->(linear->view->silu, linear->view)```, this PR update the pattern to make the linar+silu can be fused(we first need remove view ops, and then apply fusion patterns).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106300
Approved by: https://github.com/jgong5, https://github.com/jansel
